Payments were bleeding. I stopped it in a day.
Senior PM, Payments & Fraud · Rail Europe · Sep 2023–Sep 2024, then advising on payments through the 2026 launch · fraud ops team of 7
I joined during a fraud crisis. Every payment went through 3DS, fraud checks ran after capture, and nobody knew where the losses were.
- Day 1A bot attack hits one product. I switch it off the same day.
- Day 14Live Monitoring in production. The attack never came back.
- RolloutI fly to India to train the fraud team in person.
- 2+ yearsThe gains hold for over two years, until an automated fraud system takes over in 2026.
Fig. 2 · Payment acceptance by month, in points vs the average before Live Monitoring.
Read the full case
The fix. Live Monitoring was a human review layer run by the fraud operations team I trained. Acceptance rose and fraud fell together, for more than two years.
What came next. When an automated fraud system later replaced the manual process, fraud fell to almost zero but acceptance dropped back. I analysed why and set the fix plan.
The fix plan. Recalibrate the fraud model, close the gaps it created, and run a controlled A/B test instead of before-and-after comparisons.
Why 8% in the Checkout.com story? It was written early on, when acceptance was up 8%. The gains kept growing after that, reaching +14 points on average over the Live Monitoring period.